0
$\begingroup$

Let $X_1,...,X_N$ be random variables that are iid with the uniform distribution over $\mathbb S^n.$

I am curious how to see that $f(X_1,..,X_N):=\left \lvert \sum_{i=1}^N X_i \right\rvert^{-1}$ has finite expectation, i.e.

$$\int_{(\mathbb S^n)^N} f(X_1,..,X_N) \ dS(X_1)\cdots dS(X_N)<\infty$$

is finite, where $dS$ is the surface measure.

I know it is true by this answer here, which shows it for $\mathbb S^1$ and gives even the asymptotic of the integral. So I believe that if it is true on $\mathbb S^1$ it also has to be true on all other spheres, but I am looking for a more direct argument than in the above answer to see that this is indeed the case.

$\endgroup$
1
  • $\begingroup$ I suspect you want $f$ to be the inverse distance of $X_1+\cdots+X_N$ to the origin? Note that in the case $n=1$ and $N=2$ the expectation is infinite. $\endgroup$ Sep 6, 2020 at 9:37

1 Answer 1

3
$\begingroup$

Assuming you mean $f(X_1,\ldots,X_n) = |\sum_{i=1}^N X_i|^{-1}$, the case $n\geq 2$ is actually easier than $n=1$. You could use that there exists a constant $C>0$ such that for any $x\in\mathbb{R}^n$ the expectation $\mathbb{E}[ 1 / |X_1 + x| ] < C$. Then automatically also $\mathbb{E}[ 1 / |X_1 + \ldots + X_n| ] < C$ by taking $x$ to have the distribution of $X_2 + \ldots + X_N$.

Note that this argument does not work for $n=1$, because if $|x|=1$ then $\mathbb{E}[ 1 / |X_1 + x| ] = \infty$ due to a logarithmic divergence.

$\endgroup$
2
  • $\begingroup$ can the integral $\mathbb E(1/\vert X_1+x \vert)$ for $n \ge 2$ actually be worked out? $\endgroup$ Sep 6, 2020 at 11:12
  • $\begingroup$ It is not necessary for the question and it is probably not possible in elementary functions, but in terms of hypergeometric functions you should be able to work out a general expression. $\endgroup$ Sep 6, 2020 at 11:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.